CEO age and firm risk

I begin the empirical analysis by examining the relationship between CEO age and firm risk.

To do so, I first estimate Equation (2.1), in which retirement age serves as a (noisy) proxy for decreasing CEO age. The expectation is that (idiosyncratic) volatility increased significantly for the group of firms that experienced sudden deaths of retirement-age CEOs, whereas the non-retirement age group should exhibit no change in volatility. For both volatility measures, I consistently estimate two versions of the model: one that includes the additional control variables and one that does not. Table 2.5 presents the results.

Table 2.5: Effect on volatility by (non-)retirement age CEO groups

Overall Volatility Idiosyncratic Volatility

(1) (2) (3) (4)

Volatility Volatility IdioVol IdioVol Treated×Post×Non-Retirement Age CEO -0.9291 -1.0066 -0.7262 -0.8561 (0.4879) (0.4300) (0.5724) (0.4768) Treated×Post×Retirement Age CEO 3.9239* 4.3890** 3.7117* 4.0787**

(0.0544) (0.0194) (0.0747) (0.0268)

Year Fixed Effects Yes Yes Yes Yes

Firm Fixed Effects Yes Yes Yes Yes

Number of Firms 261 261 261 261

Number of Observations 1,929 1,929 1,929 1,929

Adj. R-Squared 0.2771 0.2980 0.2143 0.2445

This table presents DID analyses for the effect of CEO age on firm risk. In the regressions, the treatment effect is split for firms with sudden deaths of retirement age and non-retirement age CEOs. The dependent variables are VolatilityorIdioVol, respectively. Treated is a dummy variable equal to one for firms that experienced a sudden death and zero for the control firms. Post is a dummy variable equal to one for the four years after and zero for the four years prior to the event. Retirement Age CEOis a dummy variable equal to one for the treated firms in which the deceased were 65 or older at the time of their passing and zero otherwise. Non-Retirement Age CEO is a dummy variable equal to one for the treated firms in which the deceased were younger than 65 and zero otherwise.

Definitions for the remaining variables can be found in Table A.2 in Appendix A.2. All models include firm and year fixed effects, as well as a constant term. Thep-values are based on standard errors clustered at the firm-level and are reported in parentheses, with *, **, and *** indicating significance levels of 10%, 5%, and 1%, respectively.

As expected, the results demonstrate that the firms that experienced sudden deaths of retirement-age CEOs—that is, firms that had to decrease CEO retirement-age—experienced a significant increase in volatility, whereas those with younger deceased CEOs did not. Throughout all four models—for both overall and idiosyncratic volatility as well as with and without the additional controls—the coefficient estimates for the non-retirement age group are statistically indistinguishable from zero, whereas those for the retirement age group are positive and statistically significant at the five and ten percent levels. These results support the notion of a negative influence of CEO age on firm risk. Crucially, these results allow for a causal interpretation of the effect, because in this case, even the selection of a younger successor can be regarded as exogenous.

It is also worth mentioning that retirement age only serves as a (noisy) proxy for the age reduction.

Research has shown that in rare cases, firms hire CEOs that have passed the typical retirement age (Wang and Yin, 2020). If some of the firms that I include in the retirement age group had chosen such retirement-age successors, the effect of the decrease in age would be diluted and would thus hinder my attempt to yield significant results here. The fact that a significant effect is still identified should further support its existence. Furthermore, while this analysis provides clear advantages in terms of causal interpretation, it is more difficult to draw inferences regarding the effect’s economic magnitude.

To obtain a more detailed picture of the effect’s occurrence and its economic magnitude, I next estimate Equation (2.2), analyzing the treatment effect by the three previously defined age change groups. The expectation here is that volatility increased significantly for the group of firms that experienced a large decrease in CEO age, whereas the group that maintained CEO age stability or even increased CEO age should exhibit no change in volatility or even a decrease.

Table 2.6 illustrates the results.

The results indicate that only the group of firms with a large decrease in CEO age experienced an increase in volatility following a change in the CEO. Throughout all four models, the coefficient estimates on the interaction terms for the large decrease group are negative and (highly) statistically significant at the one or five percent levels. This corroborates the results from the previous analysis regarding the negative effect of CEO age on firm risk. Moreover, these results demonstrate that an increase in volatility is economically meaningful. Since volatility is a percental measure, the coefficient estimate of the DID interaction term for the large decrease group in Model (2), for example, indicates that, on average, stock return volatility increased by 5.0770 percentage points for these firms following the CEOs’ sudden deaths compared the control

Table 2.6: Effect on volatility by age change groups

Overall Volatility Idiosyncratic Volatility

(1) (2) (3) (4)

Volatility Volatility IdioVol IdioVol Treated×Post×Stable/Increase -1.4883 -1.4724 -0.8360 -0.9188 (0.5302) (0.5102) (0.7178) (0.6686) Treated×Post×Moderate Decrease -1.5078 -1.7430 -1.2878 -1.5752

(0.3093) (0.2384) (0.3644) (0.2684) Treated×Post×Large Decrease 4.6495** 5.0770*** 4.0505** 4.4303**

(0.0168) (0.0060) (0.0368) (0.0135)

Year Fixed Effects Yes Yes Yes Yes

Firm Fixed Effects Yes Yes Yes Yes

Number of Firms 261 261 261 261

Number of Observations 1,929 1,929 1,929 1,929

Adj. R-Squared 0.2801 0.3016 0.2161 0.2470

This table presents DID analyses for the effect of CEO age on firm risk. In the regressions, the treatment effect is split for three age change groups. The dependent variables areVolatility orIdioVol, respectively. Treated is a dummy variable equal to one for firms that experienced a sudden death and zero for the control firms. Postis a dummy variable equal to one for the four years after and zero for the four years prior to the event. Stable/Increase, Moderate Decrease, andLarge Decreaseare dummy variables equal to one for treated firms with a change in CEO age of -3 to 20, -12 to -4, and -35 to -13, respectively, and zero otherwise. Definitions for the remaining variables can be found in Table A.2 in Appendix A.2. All models include firm and year fixed effects, as well as a constant term. Thep-values are based on standard errors clustered at the firm-level and are reported in parentheses, with

*, **, and *** indicating significance levels of 10%, 5%, and 1%, respectively.

group. Considering the treated firms’ mean volatility of 39.5197 percent in the pre-event period, this implies an increase of 12.8 percent. While the coefficient estimates for idiosyncratic volatility are slightly smaller, the percental change is almost the same (i.e., 12.2 percent based on Model (4)).²⁰ Contrarily, the coefficient estimates indicating the treatment effect for the remaining two groups of firms are statistically indistinguishable from zero. This provides another interesting insight: an increase in volatility only occurs if the succeeding CEO is much younger than their predecessor.

In sum, the results presented in this section demonstrate that large decreases in CEO age lead to increased stock return volatility. This is in line with the notion formulated in Hypothesis H1 that CEO age negatively influences firm risk. Furthermore, they substantiate the firm risk results reported by Serfling (2014) and Peltomäki et al. (2020) in that they offer greater allowance for a causal interpretation. However, it remains unclear from where the increased firm risk derives, as the negative relationship is in line with both explanations described in Section 2.2. Below, I provide tests for both explanations.